The Idiosyncratic Impact of an Aggregate Shock: The Distributional Consequences of COVID-19 (IFS Working Papers)
Summary
An Institute for Fiscal Studies working paper, IFS Working Papers No. W20/15, by Michaela Benzeval and co-authors, titled The idiosyncratic impact of an aggregate shock: The distributional consequences of COVID-19 and dated May 31st 2020. It uses data from the Understanding Society COVID-19 survey collected in April 2020, linked to earlier waves of the UK household panel, to show how the economic shock of the pandemic differs across individuals. The abstract states that workers with less education and precarious employment face the biggest economic shocks, that universal credit protects those in the bottom quintile, and that opportunities for mitigation are most limited for those most in need. The paper describes its data and weighting methods and ends with tables, including Table 14 on shocks by industry.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“The Idiosyncratic Impact of an Aggregate Shock: The Distributional Consequences of COVID-19 (IFS Working Papers).”
Original at www.econstor.eu. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/2020-the-idiosyncratic-impact-of-an-aggregate-shock-the-distributional-11dbfc8685/
, SHA-256 8bf6fe1e9600145f407839c509ec4138d1ffadf7d02d7411b06be1fd90b86e74.Full text
Benzeval, Michaela et al.
Working Paper
The idiosyncratic impact of an aggregate shock: The
distributional consequences of COVID-19
IFS Working Papers, No. W20/15
Provided in Cooperation with:
The Institute for Fiscal Studies (IFS), London
Suggested Citation: Benzeval, Michaela et al. (2020) : The idiosyncratic impact of an aggregate shock:
The distributional consequences of COVID-19, IFS Working Papers, No. W20/15, Institute for Fiscal
Studies (IFS), London,
https://doi.org/10.1920/wp.ifs.2020.1520
This Version is available at:
https://hdl.handle.net/10419/223291
Standard-Nutzungsbedingungen: Terms of use:
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your personal
Zwecken und zum Privatgebrauch gespeichert und kopiert werden. and scholarly purposes.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial purposes, to
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich exhibit the documents publicly, to make them publicly available on the
machen, vertreiben oder anderweitig nutzen. internet, or to distribute or otherwise use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen If the documents have been made available under an Open Content
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, Licence (especially Creative Commons Licences), you may exercise
gelten abweichend von diesen Nutzungsbedingungen die in der dort further usage rights as specified in the indicated licence.
genannten Lizenz gewährten Nutzungsrechte.
The idiosyncratic impact of an aggregate
shock: the distributional consequences of
COVID-19
IFS Working Paper W20/15
Michaela Benzeval
Jon Burton
Thomas F. Crossley
Paul Fisher
Annette Jäckle
Hamish Low
Brendan Read
The Idiosyncratic Impact of an Aggregate Shock
The Distributional Consequences of COVID-19∗
Michaela Benzeval † Jon Burton ‡ Thomas F. Crossley§
Paul Fisher¶ Annette Jäckle k Hamish Low∗∗ Brendan Read ††
May 31st 2020
∗
Author order is alphabetical. The Understanding Society COVID-19 study is funded by the Economic
and Social Research Council (ES/K005146/1) and the Health Foundation (2076161). Fieldwork for the
survey is carried out by Ipsos MORI and Kantar. Understanding Society is an initiative funded by the
Economic and Social Research Council and various Government Departments, with scientific leadership by
the Institute for Social and Economic Research, University of Essex. The research data are distributed by
the UK Data Service. All errors are our own.
†
ISER, University of Essex and Understanding Society (email: mbenzeval@essex.ac.uk).
‡
ISER, University of Essex and Understanding Society (email: jburton@essex.ac.uk).
§
European University Institute, IFS, ESCoE and and Understanding Society (email:
thomas.crossley@EUI.eu).
¶
ISER, University of Essex and Understanding Society (email: pfishe@essex.ac.uk).
k
ISER, University of Essex and Understanding Society (email: aejack@essex.ac.uk).
∗∗
University of Oxford, IFS and Understanding Society (email: hamish.low@economics.ox.ac.uk).
††
ISER, University of Essex and Understanding Society(email: bread@essex.ac.uk).
1
Abstract
Using new data from the Understanding Society: COVID 19 survey collected in April
2020, we show how the aggregate shock caused by the pandemic affects individuals
across the distribution. The survey collects data from existing members of the Under-
standing Society panel survey who have been followed for up to 10 years. Understand-
ing society is based on probability samples and the Understanding Society Covid19
Survey is carefully constructed to support valid population inferences. Further the
panel allows comparisons with a pre-pandemic baseline. We document how the shock
of the pandemic translates into different economic shocks for different types of worker:
those with less education and precarious employment face the biggest economic shocks.
Some of those affected are able to mitigate the impact of the economic shocks: universal
credit protects those in the bottom quintile, for example. We estimate the prevalence
of the different measures individuals and households take to mitigate the shocks. We
show that the opportunities for mitigation are most limited for those most in need.
Keywords: COVID-19, job loss, inequality, mitigation, financial distress
JEL codes: C83, D31, G51, I31, J31, J63,
2
1 Introduction
The onset of COVID-19 has caused a substantial contraction in economic activity, partly
through individuals changing their behaviour in the face of the health risks and partly
because of restrictions introduced by governments. In this paper we show that the scale
of the economic impact of the COVID shock is very different for different people. These
differences arise partly because the direct shock differs depending on what sort of work
people do, and partly because individuals have differing abilities to mitigate the shock. The
aim of the paper is to highlight the reasons for the idiosyncratic nature of the economic
shocks and to show how heterogeneity in circumstances mean the same economic shock has
very different implications.
There is already a sizable literature documenting how individuals and households have
fared through the pandemic.1 Our work makes three contributions to this literature: first, our
results are based on a large, high quality survey derived from probability samples. In the UK,
“Understanding Society” has interviewed individuals annually pre-COVID, creating a panel
since 2009 (Institute for Social and Economic Research, NatCen Social Research and Kantar
Public, 2019), and additionally since the onset of COVID, individuals are being interviewed
monthly (Institute for Social and Economic Research, 2020a). The use of carefully modelled
inverse probability weights and proper probability samples is necessary to understand the
effects of an aggregate shock and avoids the ad-hoc use of calibration weights (Benzeval
et al., 2020). Second, the long panel data pre-COVID provides a clear picture of the situation
different households were in when COVID struck. In particular, we know details of their
long-run income, job security, financial fragility and economic situation, and this provides
crucial context of households situations pre-COVID. Third, to get at the question of who
is best able to mitigate the crisis, the questions post-COVID provide information on what
steps individuals are taking to mitigate losses, combined with information pre-COVID. These
three contributions enable us to provide a unique perspective on how differently COVID has
changed the economic reality faced by different households in the UK.
We split our analysis into showing differences in the economic shock; into showing differ-
ences in the ability to mitigate the economic shock; and into showing differences in outcomes
and financial security.
We find substantial heterogeneity in economic outcomes. Those who have been least
1
Among many others, on the labour market outcomes, see Adams-Prassl et al. (2020), Bell and Blanch-
flower (2020); on consumption outcomes, see Baker et al. (2020b), Baker et al. (2020a).
3
affected are a combination of those who have not experienced much of a shock and those
who despite receiving a negative shock, are able to mitigate. The first group includes those
in professional jobs able to work from home and those in industries with less-human facing
contact. The second group includes those with precarious employment at the bottom of the
income distribution but who are potentially well insured by the universal credit system. The
worst outcomes are for those experiencing severe shocks but without mechanisms to mitigate,
such as single parents, those in the lowest education groups and ethnic minorities. Further,
at this point in the crisis, the most widely used mechanisms of mitigation of earnings losses
are through self-insurance via saving and borrowing, rather than through external help.
Related Literature The immediate research on the labour market impacts of COVID
provided predictions of the likely impacts (Alon et al., 2020; Dingel and Neiman, 2020;
Yasenov, 2020). In contrast to other economic downturns, a stronger impact on women was
predicted because of the large initial negative shock to service occupations that have high
female employment shares (Alon et al., 2020); and also on industries with a concentration of
workers unable to work at home (Dingel and Neiman, 2020; Yasenov, 2020). Several studies
provide estimates of the most immediate labour market impacts of COVID (Coibion, Gorod-
nichenko and Weber, 2020; Adams-Prassl et al., 2020; Brynjolfsson et al., 2020), although
these estimates are derived from non-probability samples making population level inference
challenging. Differently, a small but growing body of evidence has exploited individual level
survey data from probability samples to estimate the labour market shocks of the early
weeks of the crisis (Béland, Brodeur and Wright, 2020; Cortes, 2020; Montenovo et al., 2020;
Mongey, Pilossoph and Weinberg, 2020). Evidence derived from the March 2020 US Current
Population Survey shows increased unemployment, decreased working hours, but little fall
in wages (Béland, Brodeur and Wright, 2020). The labour market impacts have been shown
to be bigger for men, younger workers, Hispanics, and the less educated (Cortes, 2020); al-
though (Montenovo et al., 2020) finds larger effects for women and those with larger families.
Those who cannot work remotely have also been shown to be amongst the most adversely
effected (Béland, Brodeur and Wright, 2020; Montenovo et al., 2020; Mongey, Pilossoph and
Weinberg, 2020). Larger shocks for vulnerable populations have also been documented for
the Netherlands (von Gaudecker et al., 2020) and for Norway (Alstadsæter et al., 2020); the
latter being derived from the population benefit register. A separate stream of research has
used high frequency data to study the labour market impacts of COVID: Kahn, Lange and
Wiczer (2020) (job frequency data); Bell and Blanchflower (2020) (Labour market indica-
4
tors); Watanabe and Omori (2020) (credit card transactions). Differently, the present paper
is able to track individuals over multiple pre-crisis years and so provides a fuller contex-
tual setting than does the early literature. It moves beyond the narrow measures of labour
market shocks studied above to give a more complete picture of labour market outcomes
by documenting mitigation strategies of individuals and the combined effect of shocks and
mitigation on individual’s economic outcomes.
Roadmap: The paper proceeds as follows: in section 2, we outline the data, the method
of weighting the data and the basis of population inferences. Section 3 reports the size of the
economic shocks facing different individuals. Section 4 reports the extent that individuals are
able to mitigate the impact of the shocks. Section 5 reports the overall impact on measures
of welfare.
2 Data and Methods
This paper is based on the Understanding Society COVID-19 study (Institute for Social
and Economic Research, 2020a). The Understanding Society COVID-19 study is built upon
Understanding Society: the UK Household Longitudinal Study, and uses frequent web sur-
veys to capture the experiences and behaviours of the Understanding Society participants
during the COVID-19 pandemic. This means, first, that the Understanding Society COVID-
19 study inherits the properties of Understanding Society that ensure reliable population
inferences. Second, data collected by the The Understanding Society COVID-19 study can
be linked to data collected on the same participants, and their households, in past waves of
the Main study (and it will, in the future, be possible to link them to future waves as well,
opening up the possibility of tracking longer term changes.) The current paper employs data
from the first wave of the Understanding Society COVID-19 study, which was fielded in late
April 2020, alongside contextual information from past waves of Understanding Society.
2.1 Understanding Society
Understanding Society (University of Essex Institute for Social and Economic Research,
NatCen Social Research and Kantar Public, n.d.) is the UK’s main longitudinal Household
Survey, and one of the largest household panel studies in the world. It began in 2009 but
carries on from (and incorporates the sample of) the earlier British Household Panel survey
which ran from 1991 to 2008. Understanding Society attempts to interview all adults in sam-
5
ple households annually. Understanding Society began with entirely face-to-face interviews
but has transitioned so that it is a mixed mode design, with some panel members responding
via a face-to-face interview and some completing a web interview.
Understanding Society comprises four distinct samples.2 Each began as a probability
sample. All samples other than the Northern Ireland sample had a clustered and stratified
design.
A key feature of probability samples is every unit in the target population has a knowable,
nonzero probability of selection (Valliant and Dever, 2018). This offers two advantages
over other types of samples (such as convenience or quota samples). First, the fact that
all units in the target population have a nonzero probability of selections ensures that,
with sufficiently large sample sizes, the full range of heterogeneity in the target population
will be captured. Second, known selection probabilities mean that consistent estimates
of population parameters and associated inferences can be obtained with well established
statistical methods involving inverse-probability weighting.
Of course, real samples deviate form the theoretical ideal of a probability sample because
of non-response, including, in the case of longitudinal studies, attrition. It is, nevertheless,
of great advantage to begin from probability samples. Other types of samples may have a
zero probability that certain parts of the target population will enter the sample. Second,
while statistical adjustments may be needed to account for nonrandom non-response and
attrition, such adjustments have less work to do if the initial selection probabilities are
known. A further point is that when a study begins with a probability sample, information is
typically available on non-respondents. This is particularly true in longitudinal studies where
rich information on individuals who attrit is available from past waves of the survey. Such
information is obviously of great value in modeling response and attrition. This contrasts
with convenience or quota samples where information is only available for respondents, and
differences between respondents and non-respondents can only be inferred indirectly.
Understanding Society makes considerable efforts, and employs state of the art methods,
to minimize non-response and attrition. At the same time, it provides weights to account
for the nonrandom nature of the residual unavoidable attrition. These inverse-probability
weights are based on very carefully modeling of response and attrition. Although the proba-
bility that a given individual is included in the sample at a given wave is not known ex ante, it
can be very credibly estimated, combining the best available statistical methods in conjunc-
2
These are: the General Population Sample, The Ethnic Minority Boost Sample, The Immigrant and
Ethnic Minority Boost sample, and the former British Household Panel Survey sample.
6
tion the rich information available from past waves. The development of the Understanding
Society weights is described in (Lynn and Kaminska, 2010).
The extent to which Understanding Society is “representative”, in the sense of supporting
high quality inferences about population quantities, is continually evaluated: see Benzeval
et al. (2020) and the references therein. The study has been repeatedly judged to be of high
quality. As just one example, Understanding Society income data aligns well with national
statistics on the income distribution in the UK (Fisher et al., 2019).
2.2 The Understanding Society COVID-19 Web Survey
The first wave of the Understanding Society COVID-19 Study was fielded between April
24 and 29 2020. The study was issued to all members of the four Understanding Society
samples who were aged sixteen or over in April 2020 and who belonged to active households
(one that had participated in at least one of the last two waves of the main Understanding
Society study).3
Pre-notification letters introducing the study were sent to sample members on 17 April.
Respondents were offered a small financial incentive for each wave of the Understanding
Society COVID-19 Study 4 . Invitations to the survey were then sent by email and/or SMS
text message, or by post. Reminders were sent on days 2, 3, and 6 of the seven day fieldwork
period.
The first wave web questionnaire took approximate 20 minute to complete. In addition
to economics and household finances content analyzed in this paper, the questionnaire asked
questions about health, health behaviours and home schooling. Further information can be
found in Institute for Social and Economic Research (2020b).
2.3 Population Inferences
Economists often eschew the use of survey weights (both design weights and non-response
adjustments). The justification that is that where the object is to estimate a correctly spec-
ified model, weights may have no benefits, and come at cost of lost efficiency (Moffitt,
Fitzgerald and Gottschalk, 1999). However, in a distributional analysis such as the one un-
3
There were some minor exceptions, including individuals who were adamant refusals to the main study,
who had a foreign address.
4
Respondents were offered a 2 pound incentive for each monthly survey, which they could accumulate
and exchange for a range of gift cards and vouchers.
7
dertaken in this paper, this argument does not apply. The objective is to estimate population
statistics, rather than an economic model. As a consequence, it is of utmost important to
use survey weights to ensure consistent estimates of those statistics.
Given the robust evidence that (suitably weighted) waves of the Understanding Society
main survey provide reliable population evidence, the key consideration is non-response to
Understanding Society COVID-19 Study among respondents to the last complete wave of
the main survey (Wave 9). Among those who had given a full adult interview in the Wave 9
annual interview, the response rate to to the COVID-19 Wave 1 Survey was 48.6%.5 . This
is a very good response rate for a voluntary web survey when such surveys attempt to reach
a target list (convenience and quota samples do not have a knowable response rate). It is
also close to the response rate of large government surveys in the UK.6 Nevertheless, this is
significantly below the 85-90 % overall wave-on-wave retention rate that the Understanding
Society main survey achieves by following up web non-respondents by direct interviewer
contact. 7
Cross-sectional individual weights are provided with the Understanding Society COVID-
19 Wave 1 data.8 These inverse-probability weights were created via an adjustment to the
cross-sectional weights available for Wave 9 of the main survey. This means that probability
of response to the COVID-19 Wave 1 Survey is modeled as the product of the probability of
COVID-19 Wave 1 response conditional on main survey Wave 9 response and the probability
of Wave 9 response. The conditional probability of COVID-19 Wave 1 response is modeled by
logistic regression, with a step-wise variable selection. The choice set of predictors include
basic demographics, household composition, economic variables and health variables. In
addition, both econometrics and survey statistics literature emphasize the importance of
including in weighting models variables that predict response and potentially correlated with
outcomes being studied, but unlikely to be including in standard economic or social science
models. Key variables sources of such variables are previous survey outcomes, survey design
variables and survey para data. Several such variables turn out to be good predictors of
the probability of COVID-19 Wave 1 response conditional on main survey Wave 9 response,
and are included in the weighting models. These include indicator variables for the types
5
or 46.0% if partial interviews are excluded
6
For example, the Labour Force Survey - to which many web surveys with quota or convenience samples
calibrate - has response rate of about 55% at the first wave, falling with subsequent and about 40 % overall.
The Family Resources Survey which is the basis for official income statistics had a response rate of 52% in
2017/18.
7
In the main survey, the individual-level online response rate is 50%-55% (of those invited).
8
The weights were created by the authors of this paper
8
of contact information the survey team held about the respondent prior to the COVID-19
Wave 1 survey (email address, mobile phone number, both, neither) and the realized mode
of previous waves of the main survey (recall that the Understanding Society main survey is
multi-mode). The former may affect the salience of the survey request while the latter may
be related to how easy the respondent would find it to complete a web survey. Either could
quite plausibly be related to whether the respondent is employed or the kind work they do.
The final Understanding Society COVID-19 Wave 1 cross-sectional weights are calculated
as the inverse of the estimated response propensity, and bounded at 3.5 times the median to
control variability.
It is worth contrasting these weights with the ex post calibration weights that are often
produced for other web surveys with convenience or quota samples. Understanding Society
COVID-19 Wave 1 survey weights are based on a rich set of covariates, and on information
both respondents and non-respondents (or non-attritors). Ex post calibration weights are
typically based on a small set of covariates (age, education and gender is common) and
typically no information is available on non-respondents, so that response or retention prob-
abilities cannot be estimated directly. Instead, necessary adjustments are inferred indirectly
by comparison to external sources; those sources are often surveys, such as the Labour Force
Survey, which have quality markers (such as response rates) comparable to Understanding
Society.
2.4 Two Tests for Attrition Bias
Given the consistent estimates of population statistics in a distributional analysis such as
the one we present before, it is important to assess the extent to which the Understanding
Society COVID-19 weights deal with nonrandom attrition from Wave 9 of the main study.
We implemented two types of test.
As noted above, non-response to the COVID-19 Wave 1 Survey, conditional on response
to Wave 9 of the main survey, can be viewed as attrition between the two. The econometrics
literature contains several suggestions for tests of nonrandom attrition in panel data (Beck-
etti et al., 1988; Fitzgerald, Gottschalk and Moffitt, 1998), where typically the test is for
whether attrition is nonrandom with respect to a lagged outcome, conditional on some set
of (regression) model covariates, and without using weights. These are not suitable for our
purposes, where the focus is on distributional population statistics, and where weights will
be employed. We therefore propose a straight-forward test of weights in panel data which
9
we have not been able to find in the previous literature.
Consider two waves, t and t + k of a panel survey. In our application these will be Wave 9
of the main Understanding Society survey and Wave 1 of the COVID-19 survey. Let i index
wave t respondents and Nt the total number of such respondents (the wave t sample size).
Let Y be some variable of interest and µY tr be the rth uncentered population moment of Y
at the time of wave t which we assume exists and is finite (when r = 1, this is the population
mean). Then let Yt,i be the observation of Y for individual i in wave t. Let Rt+k,i indicate
wave t + k response. Among wave t respondents, Rt+k,i = 1 if the individual responds to
wave t + k (that is, does not attrit) and 0 otherwise. Finally, let wti and wt+k,i be the wave
t and t + k cross-sectional survey weights.
Under the null hypothesis that wti and wt+k,i are, respectively, the inverse response prob-
abilities for waves t and t + k, the following are both consistent estimators of µr :
PNt r
i=1 wt,i Yt,i
PNt (1)
i=1 wt,i
PNt r
i=1 Rt+k,i wt+k,i Yt,i
PNt (2)
i=1 Rt+k,i wt+k,i
Note however, that estimator (1) is more efficient. This gives us a natural ”Hausman-
type” test of the null hypothesis based on the difference between (1) and (2) (divided by the
appropriate variance of the difference.) This test can be implemented for any Y (and for
different values of r, though of course if Y is a binary variable, only r = 1 is of interest.)
Table 6 reports the results of this test for a selection of Understanding Society Main Survey
Wave 9 variables, using Wave 9 respondents and associated weights for (1) and COVID-19
web survey respondents and associated weights for (2). The differences between (1) and (2)
and the associated p-values are reported in the final column of Table 6.
For comparison purposes, we also constructed a “basic” set of weights in which the proba-
bility of COVID-19 Wave 1 response conditional on main survey Wave 9 response is modeled
as a function of just age, gender and education. These illustrate the advantage of the rich
background and survey information that is used to model that probability in the actual
(”refined”) COVID-19 Wave 1 weights (though note that even the basic weights we create
here benefit from the fact that we have direct information on non-respondents). The parallel
test based on (1) and (2) using these basic weights is reported in the penultimate column of
Table 6.
10
Unsurprisingly, both the refined and basic COVID-19 Wave 1 weights lead to values of
estimate (2) which are close to estimate (1) (based on Wave 9 Main Survey respondents
and weights) if Y is chosen from the variables included in the basic weight model (age,
gender and education). For brevity, these tests are omitted from the Table. The variables
Y reported in Table 6 are divided into two groups. Those that are included in the refined
weight model (but not the basic weight) and those that are not in either of the probability
models that underpin the basic and refined weights. Table 6 shows, first, that with the
actual (refined) COVID-19 Wave 1 weights, for most variables we are unable to reject the
null hypothesis (that these weights and the Wave 9 main stage weights both capture the
probability of response for their respective samples). The COVID-19 Wave 1 sample and
associated weights does seem to lead to an mean income estimate at Wave 9 that is a bit too
high, and, conversely, a probability of core benefit receipt which is a bit too low. The second
thing that Table 6 demonstrates is that that refined weights improve on the basic weights,
and not just for variables included in the weighting model.
Because the COVID-19 Wave 1 weights are based on prior wave observables, they cannot
correct for selection into response based on contemporaneous shocks (this is of course also
true of weights based on permanent characteristics such as age, education and gender).
Similarly, the tests presented in Table 6, which are based on prior observables, cannot test
for selection on contemporaneous shocks. This may be of particular concern during the
COVID-19 pandemic. Response is plausibly related to time demands, and could be related
to very recent shocks to employment, hours, or caring or home-schooling responsibility. The
ability and willingness to complete a survey may also be related to health shocks. This would
mean that survey non-response was not missing at random (NMAR) with respect to past
observables. It is possible to test for this if an instrument for response is available. That
is, one requires a variable Z, that predicts response but is uncorrelated with outcomes of
interest. Unusually, in the COVID-19 Wave 1 survey we tried to generate such a variable.
In particular, we randomly assigned potential respondents to “batches” and varied the time
when batches were invited to take the survey across the first day of the seven-day field work
period. This means that early batches had almost an additional day to complete the survey
relative to the latest batches, and also that the invitation (for example email or SMS message)
was first received at a different time of day. This turns out to have good predictive power for
response. Note that despite the batch number being a useful predictor of response, it is not
a variable to include in weighting models. This is because the randomization of assignment
11
to batch implies that it is uncorrelated with any outcomes of interest Y , including at time
t + k. If the outcome Y is not uncorrelated with Z, in the sample of respondents, this is
evidence of selection.
We have implemented this test in all the regressions we run the selection instrument is
never significant at conventional levels in those regressions. Thus this instrument does not
lead us to reject the null hypothesis of no selection on contemporaneous shocks.
2.5 Analysis Sample and Additional Methods
The analysis that follows is based on the subset of respondents aged 20 to 65, and so
capture the UK population of this age range.
The underlying Understanding Society samples are clustered unit (PSU) and stratified
random samples, and so the COVID-19 sample inherits this structure. We use the SVYSET
suite of commands in STATA to appropriately adjust standard errors for the resulting design
effects.
3 Labour Market Shocks
In this section, we show results on the extent of the labour market shocks that individuals
face and the reasons for these shocks.
We describe labour market status using four measures: whether an individual is employed
and whether they are working a positive number of hours; and for those who are employed,
average hours worked and average earnings. In Table 1, Table 2 and Figure 1, we show mean
values for these four measures using reports from February 2020 and April 2020. Tables 1 and
2 and Figure 1 show variation in mean values by individual characteristics. Table 3 shows
the changes in the measures using regression to identify the marginal effects of individual
characteristics. We allocate individuals to quintiles of “long run income” where the 3 waves
prior to the start of COVID-19 are averaged to define long run income. Income includes
earned and unearned income, net of tax and inclusive of any benefits received, equivalised
by household composition.
Table 1 and Figure 1 show that across all individuals, employment has held up well. On
the other hand, there was a significant fall in the fraction working positive hours: by the
end of April only 56% of working age individuals were working positive hours, down from
almost 80% in February. This is not surprising given the Job Protection and Retention
12
(furlough) scheme introduced in March 2020.9 However, this difference between employment
and working-positive-hours does highlight starkly the potential unemployment problem in
the coming months. The fall in the fraction working positive hours and in average hours
worked was particularly acute for those educated to less than degree level and for single
parents, and for those in lower long-run income quintiles. We verify these unconditional
mean effects through multivariate regressions of the changes reported in Table 3.
Table 2 shows the variation in mean values of labour market outcomes by the character-
istics of the job, including contract type and occupation. This table uses occupation reports
from the wave 9 survey in 2017/18 and conditions on being employed in February 2020. In
the Appendix, in Table 14, we show the variation by industry. We split job characteristics
according to how hours of work are set, 10 by whether the individual worked from home at
all prior to February, and by the occupation. Here the heterogeneity in the extent of labour
market shocks is striking: hardest hit are those individuals where the employer does not
guarantee any minimum number of hours, with the fraction of those employed in February
that work positive hours has fallen from 97% to 41%. Similarly badly hit are those who
never worked at home prior to the crisis; and those in industries which involve contact with
people. In “elementary” occupations11 , the fall in those working positive hours is from 98%
to 51%, whereas in “professional” occupations, the fall is only from 99% to 85%. The split
by industry shown in Table 14 in the Appendix reinforces this picture: substantial falls in
hours worked are in food service and construction, with minimal falls in finance.
Table 4 shows reported reasons for the fall in hours worked for those who have experienced
a decline. It may be caused directly by the health shock, or indirectly by restrictions in the
economy in the face of the health shock, or for non-health related reasons. The key point to
take from Table 4 is that the decline in hours is driven by the economic restrictions. Over
43% of those reporting a decline in hours were furloughed, and this is even more prevalent
among those who have never worked at home. A further 14% of all those experiencing a
decline cite the loss of self-employment business either due directly to restrictions or due to
reductions in demand. Among the self-employed, this rises to 61%. By contrast, only 7%
report health as a reason for the decline in hours, and 7% report caring for others. For some
9
This scheme allowed workers to be ”furloughed” by their firms, which meant 80% of workers pay would
be covered by a government subsidy, conditional on the worker not actually providing any hours of work.
10
We additionally have information on how payment is determined, but this is closely correlated with how
hours are set.
11
Occupation codes are those used by the ONS and described at https://onsdigital.github.io/dp-
classification-tools/standard-occupational-classification/ONS SOC hierarchy view.html
13
households, however, these health and caring reasons are much more important: caring for
others is capturing those with children needing caring; and health restricting work has a
steep age gradient.
A further point to draw from Table 4 is the differential prevalence of furlough and of
unemployment by ethnicity. BAME individuals are 14 percentage points less likely to be
furloughed and 13 percentage points more likely to be unemployed than non-BAME individ-
uals.
4 Mechanisms of Mitigation
Section 3 showed that the effect of the COVID-19 crisis on labour market outcomes varies
substantially across individuals. These individuals differ also in their ability to mitigate these
shocks. In this section, we start by considering how the labour market shocks translate into a
change in net household earnings. We then show how individuals have been able to mitigate
these losses, distinguishing between using self-insurance, such as running down saving or
borrowing, and using external sources, such as universal credit and transfers from friends
and family.
The measure of earnings we use is net, equivalised weekly household earnings of the in-
dividual respondents, including earnings from employment and self-employment. We use an
inverse-hyperbolic sine transformation, as in Burbidge, Magee and Robb (1988), to transform
earnings changes into growth rates.
Figure 2 shows the level of household earnings in February and April by long-run income
quintile. We further break the earnings change down by characteristics of the individual
in Table 5. The figure highlights that the decline in household earnings is seen across the
distribution, although the proportional decline is greatest for the bottom quintile. The fall
in average household earnings is 8%, and 23% of individuals reported a loss of more than
20% of household earnings. The loss of household earnings was particularly severe for single
parents, with almost a third facing a decline of over 20%.
In Table 6 we further explore the distribution of household earnings losses within quin-
tile. For each quintile of long-run income, we report the proportional chance in household
income at the 10th, 20th, 25th, 50th, 75th and 90th percentiles. Within each quintile, the
median individual experiences no change in household earnings. However, within each of
the quintiles, there is a substantial decline at the 10th percentile: even in the top quintile,
14
the proportional change is a decline of 43%. However, as shown earlier, the bottom quintiles
have the largest declines and the highest fraction of people experiencing declines.
These decline in household earnings do not necessarily translate into declines in house-
hold income or living standards, partly because households have other sources of income,
such as universal credit, but also because different households have different possibilities of
mitigation.
Figure 3 summarizes actions taken to mitigate earnings losses by individuals who have
reported a decline in household earnings. We split methods of mitigation into self-insurance,
which are changes within the household through saving, borrowing or additional work, and
into external sources, which includes transfers from other family of friends as well as state
benefits. At this stage in the crisis, individuals are relying more on self-insurance, and in
particular on their own savings, than external help to mitigate losses: more than a quar-
ter have drawn down their savings. Significant numbers have also increased borrowing or
asked for a mortgage holiday. However, different individuals have used different mechanisms:
multiple adult households have relied more on savings compared to single parents, whereas
single parents have relied more on borrowing compared to multiple adult families. Further,
in terms of differences by ethnicity, BAME individuals are more than twice as likely to have
resorted to borrowing than non-BAME individuals.
External support is more widespread among the bottom quintiles, with transfers from
friends or family somewhat larger than increased applications for Universal Credit. This
financial assistance from friends and family is much more common for individuals within the
lowest quintile of household income, for the young, and for single parents.
5 Economic Outcomes
The variation in labour market shocks and their implications for household earnings, and
the variation in mechanisms of mitigation leave households in very different financial situa-
tions. We would typically use changes in consumption for the different households to capture
the loss in standard of living associated with the shocks. In this crisis, however, the change
in consumption may be misleading as a measure of welfare loss because of changes in supply:
many usual spending categories, such as spending on food out of the home, have been sub-
stantially reduced by government restrictions. On the other hand, some of the consumption
falls will reflect reduced demand in the face of earnings losses. We start this section on the
15
economic outcomes for households by showing the heterogeneity in consumption changes.
We then turn to more direct measures of welfare loss, looking first at the extent that indi-
viduals report having fallen into arrears, and then at subjective assessments of their financial
situation. We end the section by showing expectations of their future situation.
Figure 4 shows the fraction of individuals who report cutting their spending since February
2020, by income quintile. For those whose household earnings have fallen by at least 5%, more
than two-thirds report cutting their spending. This fall in spending reflects the insufficiency
of mechanisms of mitigation to prevent a loss in living standards. However, almost 20%
of those with no loss in household earnings (or even an increase) also report a cut in their
spending, and this must reflect either anticipating hard times to follow or a reduction in
opportunities to spend. To the extent that falls in consumption are simply reductions in
opportunities to spend, this suggests a build up of demand that would generate a rebound.
To the extent that falls reflect necessary cut backs, there is likely to be more persistent
demand shortages.
To further identify the extent that the different sources of mitigation have not protected
living standards, we report in Table 8 and Figure 5 the incidence of financial arrears. We
show financial arrears by showing the fraction of respondents reporting that their household
was behind with housing payments (mortgage or rent), and the fraction behind with other
bills. We compare here to wave 9 of the main survey, which occurred in 2017-18. Table 8
also reports the fraction of respondents reporting that their household was hungry but did
not eat at some time in the last week.
Overall, there is an increase in the fraction behind with bills from 5.2% to 7.4%. But this
overall average masks the heterogeneity: the fraction who are behind with bills has risen by
5 percentage points (and from a high base) for single parents, and similarly for those in the
bottom quintile, those with the lowest education and for BAME individuals. However, in a
multivariate regression shown in Table 10, the conditional effects are insignificant.
In terms of arrears in housing, there is no significant difference in the fraction who report
now being behind with housing. Where the increase in housing arrears shows up is in the
highest two quintiles, but this is from a very low level. For the bottom quintiles, housing
arrears are high but there is no evidence of the situation worsening. By contrast, BAME
individuals are now 4.3 percentage points more likely to report being behind with housing.
The size of this unconditional difference remains when we calculate conditional effects using
regression, but the conditional effect is not statistically significant.
16
The final column of Table 8 shows the fraction reporting a decline in spending by indi-
vidual characteristics. The fraction is particularly high for BAME individuals, households
with children, and those at the bottom of the income distribution.
An alternative measure of the consequences of the crisis is to use subjective measures of
individual’s financial situations. Table 9 reports the subjective financial situation of indi-
viduals as they assess it now and compared to 2017-2018. Table 11 reports how individuals
expect their financial situation to change in the next month.
When we compare how each individual perceives their financial situation now with how
they perceived it in 2017-2018, households are not reporting a worsening situation, and if
anything the situation has improved. These comparisons are for the same set of individuals
in each period. However, the anchoring of this question into the aggregate situation may
make comparisons difficult as the financial situation may lose salience compared to concerns
about health.
Further, while we show evidence that the current situation is not perceived to be worse
than the past, Table 11 shows how individuals expect this to change by the end of May.
Overall, more than twice as many people expect their financial situation to get worse as
those who expect it to get better over this month. Further, those who we have shown to be
in the worst situation currently are also those with the most concerns going forward: the
fraction expecting their situation to get worse rises to two and a half times those expecting an
improvement for BAME individuals, to three times for those in the bottom income quintile,
and to more than three times for single parents.
6 Conclusions
The COVID-19 pandemic has clearly resulted in a substantial aggregate economic shock:
countries all around the world are affected and within country, all individuals are affected.
This paper shows that the aggregate effects mask considerable differences in which individuals
are affected and considerable differences in how individuals are able to mitigate the effects.
We use new high-quality UK data derived from probability samples: the Understanding
Society COVID-19 Survey. These data confirm that impacts are hugely heterogeneous. One
month after the “stay at home” policy was introduced, 60% of individuals in the UK had
had essentially no loss of household earnings (either no loss or a loss of less than 5%). At
the same time, 23% reported household income had fallen by more than 20%. Part of this
17
limited impact on household earnings is due to government furlough scheme that protected
employment alongside requiring furloughed individuals to work zero hours. We see this in
the data: the fall in employment is only two percentage points, whereas the fraction working
positive hours has fallen by 24 percentage points.
We show how these economic losses differ across individual characteristics and the char-
acteristics of jobs that people do. The economic losses have been highly regressive. The
welfare cost of the economic shocks depend both on the size of these direct shocks and also
on the resources and mechanisms households have to mitigate the shocks. We show that, to
date, self-insurance through using savings and additional borrowing is more prevalent than
external support from universal credit and transfers from friends and family. Further, the
largest economic shocks have fallen on those least able to mitigate. Those most affected are
BAME individuals, single parents and those in the lowest quintile of long-run income.
Future waves of the Understanding Society COVID-19 survey will enable an assessment
of how this striking heterogeneity across individuals evolves as the crisis proceeds.
18
References
Adams-Prassl, Abi, Teodora Boneva, Marta Golin, and Christopher Rauh. 2020.
“Inequality in the Impact of the Coronavirus Shock: Evidence from Real Time Surveys.”
Institute of Labor Economics (IZA) IZA Discussion Papers 13183.
Alon, Titan M, Matthias Doepke, Jane Olmstead-Rumsey, and Michèle Tertilt.
2020. “The Impact of COVID-19 on Gender Equality.” National Bureau of Economic
Research Working Paper 26947.
Alstadsæter, Annette, Bernt Bratsberg, Gaute Eielsen, Wojciech Kopczuk,
Simen Markussen, Oddbjorn Raaum, and Knut Røed. 2020. “The First Weeks
of the Coronavirus Crisis: Who Got Hit, When and Why? Evidence from Norway.” Na-
tional Bureau of Economic Research Working Paper 27131.
Baker, Scott R, R.A. Farrokhnia, Steffen Meyer, Michaela Pagel, and Constan-
tine Yannelis. 2020a. “How Does Household Spending Respond to an Epidemic? Con-
sumption During the 2020 COVID-19 Pandemic.” National Bureau of Economic Research
Working Paper 26949.
Baker, Scott R, R. A Farrokhnia, Steffen Meyer, Michaela Pagel, and Constan-
tine Yannelis. 2020b. “Income, Liquidity, and the Consumption Response to the 2020
Economic Stimulus Payments.” National Bureau of Economic Research Working Paper
27097.
Becketti, Sean, William Gould, Lee Lillard, and Finis Welch. 1988. “The Panel
Study of Income Dynamics after Fourteen Years: An Evaluation.” Journal of Labor Eco-
nomics, 6(4): 472–492.
Béland, Louis-Philippe, Abel Brodeur, and Taylor Wright. 2020. “The short-term
economic consequences of Covid-19: exposure to disease, remote work and government
response.”
Bell, David, and David Blanchflower. 2020. “US and UK Labour Markets Before and
During the COVID-19 Crash.” National Institute Economic Review, 252: 52 – 69.
Benzeval, Michaela, Christopher R. Bollinger, Jonathan Burton, Thomas F.
Crossley, and Peter Lynn. 2020. “The representativeness of Understanding Society.” ,
(2020-08).
Brynjolfsson, Erik, John Horton, Adam Ozimek, Daniel Rock, Garima Sharma,
and Hong Yi Tu Ye. 2020. “COVID-19 and Remote Work: An Early Look at US
Data.” Mimeo, available at http://ide.mit.edu/publications/covid-19-and-remote-work-
early-look-us-data.
Burbidge, John B., Lonnie Magee, and A. Leslie Robb. 1988. “Alternative Transfor-
mations to Handle Extreme Values of the Dependent Variable.” Journal of the American
Statistical Association, 83(401): 123–127.
19
Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber. 2020. “Labor Markets
during the COVID-19 Crisis:A Preliminary View.” IZA Institute of Labor Economics
Working Paper 13139.
Cortes, Guido Matias. 2020. “Heterogeneous Labor Market Impacts During the Early
Stages of the Covid-19 Pandemic.” Rimini Centre for Economic Analysis Working Paper
series 20-13.
Dingel, Jonathan I, and Brent Neiman. 2020. “How Many Jobs Can be Done at Home?”
National Bureau of Economic Research Working Paper 26948.
Fisher, Paul, Laura Fumagalli, Nick Buck, and Silvia Avram. 2019. “Understanding
Society and its income data.” , (2019-08).
Fitzgerald, John, Peter Gottschalk, and Robert Moffitt. 1998. “An Analysis of
Sample Attrition in Panel Data: The Michigan Panel Study of Income Dynamics.” Journal
of Human Resources, 33(2): 251–299.
Institute for Social and Economic Research. 2020a. Understanding Society COVID-19
Study, 2020. [data collection]. 1st Edition. UK Data Service. SN: 8644, 10.5255/UKDA-
SN-8644-1.
Institute for Social and Economic Research. 2020b. Understanding Society COVID-19
User Guide. Version 1.0. University of Essex, Colchester.
Institute for Social and Economic Research, NatCen Social Research, and Kantar
Public. 2019. “Understanding Society: Waves 1-9, 2009-2018 and Harmonised BHPS:
Waves 1-18, 1991-2009.” [data collection]. 12th Edition. UK Data Service. SN: 6614.
Kahn, Lisa B, Fabian Lange, and David G Wiczer. 2020. “Labor Demand in the
Time of COVID-19: Evidence from Vacancy Postings and UI Claims.” National Bureau
of Economic Research Working Paper 27061.
Lynn, Peter, and Olena Kaminska. 2010. “Weighting strategy for Understanding Soci-
ety.” , (2010-05).
Moffitt, Robert, John Fitzgerald, and Peter Gottschalk. 1999. “Sample Attrition in
Panel Data: The Role of Selection on Observables.” Annals of Economics and Statistics,
, (55-56): 129–152.
Mongey, Simon, Laura Pilossoph, and Alex Weinberg. 2020. “Which Workers Bear
the Burden of Social Distancing Policies?” National Bureau of Economic Research Work-
ing Paper 27085.
Montenovo, Laura, Xuan Jiang, Felipe Lozano Rojas, Ian M Schmutte, Kosali I
Simon, Bruce A Weinberg, and Coady Wing. 2020. “Determinants of Disparities in
Covid-19 Job Losses.” National Bureau of Economic Research Working Paper 27132.
University of Essex Institute for Social and Economic Research, NatCen Social
Research, and Kantar Public. n.d.. “Understanding Society: Waves 1-9, 2009-2018
and Harmonised BHPS: Waves 1-18, 1991-2009. [data collection]. 12th Edition.”
20
Valliant, Richard, and Jill Dever. 2018. Survey weights: a step-by-step guide to calcula-
tion. Stata Press College Station, TX.
von Gaudecker, Hans-Martin, Radost Holler, Lena Janys, Bettina Siflinger, and
Christian Zimpelmann. 2020. “Labour Supply in the Early Stages of the COVID-19
Pandemic: Empirical Evidence on Hours, Home Office, and Expectations.” IZA Institute
of Labor Economics Working Paper 13158.
Watanabe, Tsutomu, and Yuki Omori. 2020. “How Much Did People Refrain from
Service Consumption due to the Outbreak of COVID-19?” Center for Advanced Research
in Finance, Faculty of Economics, The University of Tokyo CARF F-Series 477.
Yasenov, Vasil. 2020. “Who Can Work from Home?” IZA Institute of Labor Economics
Working Paper 13197.
21
Figures and Tables
22
Table 1: Labour Market Shocks by Individual Characteristics
Positive Positive
Employed Employed hours hours Hours Hours Earnings Earnings
(Feb) (April) (Feb) (April) (Feb) (April) (Feb) (April)
All 80.4 78.4 79.5 55.8 34.7 23.0 429.0 380.6
Gender:
Men 84.5 82.3 83.8 59.5 38.6 25.5 506.1 439.1
Women 76.9 75.1 75.9 52.7 31.0 20.6 357.4 326.2
Ethnicity:
Not BAME 81.1 79.4 80.4 56.2 34.8 22.9 429.7 382.5
BAME 72.0 67.4 70.6 50.9 33.7 23.5 421.4 358.5
Age:
Age 20-29 79.8 75.1 78.4 48.2 33.5 20.3 326.0 290.2
Age 30-39 86.8 85.1 85.7 60.9 35.4 23.6 448.2 394.1
Age 40-49 86.2 85.8 85.7 65.2 35.9 25.7 484.2 439.2
Age 50-59 81.8 80.4 81.3 58.7 35.0 23.5 451.8 401.6
Age 60-65 58.8 56.4 57.9 36.8 31.4 18.4 376.2 313.0
23 Education:
GCSE or lower 72.6 70.0 71.7 43.6 34.1 19.6 332.1 284.5
A-level 79.5 76.9 78.8 49.7 34.1 20.3 363.9 316.0
Degree 85.8 84.4 84.9 66.7 35.3 26.0 509.6 461.4
Household type:
Single adult, no children 71.5 70.4 70.9 51.5 35.8 24.4 427.3 385.7
Single adult, children 74.6 74.2 74.3 48.8 32.1 18.8 399.6 319.3
Multiple adult, no children 78.6 75.8 77.8 54.2 34.9 23.2 414.7 368.2
Multiple adult, children 85.2 83.9 84.2 59.4 34.4 22.7 448.6 399.0
Long-run income quintile:
1 64.8 62.0 64.0 39.1 31.6 17.8 291.4 240.1
2 78.9 76.2 78.2 49.2 33.9 20.2 335.9 293.6
3 85.2 84.0 84.7 57.8 35.3 22.5 381.1 355.7
4 86.8 85.1 86.3 64.5 35.8 25.6 464.5 412.1
5 85.4 83.8 83.8 67.4 35.9 27.1 620.2 549.6
Notes: Employment, hours and earnings include both employees and self-employees. Columns 1-4 are population percentages and refer to all individuals;
columns 5-8 are weekly means and refer to those employed in February. Sample sizes are: 10,803 (col 1-4), 8745 (col 5-6), and 8015 (col 7-8).
Table 2: Labour Market Shocks by Job Characteristics
Positive Positive
Employed hours hours Hours Hours Earnings Earnings
(April) (Feb) (April) (Feb) (April) (Feb) (April)
Worker type:
Fixed hours 97.1 99.4 72.4 35.0 24.7 433.7 407.0
Flexible hours 96.2 97.1 74.2 35.9 27.6 522.1 490.2
Employer sets (sure min.) 94.7 98.4 58.9 34.2 20.3 346.4 308.1
Employer sets (no sure min.) 85.7 96.5 41.2 32.5 14.9 350.0 277.5
Self-employed 94.9 98.8 59.2 33.3 15.8 417.7 256.1
Works at home (Feb):
Sometimes or always 97.6 99.1 78.6 35.5 26.6 548.4 485.1
24 Never 95.6 98.9 64.2 34.3 21.4 376.6 334.7
Elementary 90.7 98.0 50.9 31.4 15.6 264.7 212.9
Skilled trades 96.8 99.7 55.0 40.0 18.6 416.1 318.7
Sales, customer service 94.4 99.0 55.2 30.4 17.1 269.1 246.0
Process, plant, machine operatives 96.4 99.3 61.1 40.0 24.4 389.6 328.0
Caring, leisure, other service 97.2 98.8 63.8 31.1 19.2 284.0 266.8
Administrative, secretarial 96.5 99.4 71.5 32.7 23.1 346.1 307.5
Managers, directors 97.0 99.4 72.9 39.3 27.3 608.5 548.4
Associate professional, technical 98.1 99.2 77.4 35.4 26.3 484.3 445.4
Professional 97.3 98.5 84.7 36.0 29.3 554.1 512.4
Notes: See table 1 notes. The sample is individuals in employment in February 2020 who reported a wave 9 occupation. This is 5613 individuals. The fraction employed in February
is therefore 100%. Occupation is ordered by column 4 (fraction working positive hours in April) and is collected at the wave 9 (2017-18) interviews.
Table 3: Changes in Labour Market Status
Employed Positive hours Hours Earnings
Constant -0.03 -0.23∗∗∗ -12.08∗∗∗ -0.02
(0.02) (0.04) (1.63) (0.01)
Gender:
Male ref ref ref ref
Women 0.00 0.01 2.78∗∗∗ 0.00
(0.01) (0.01) (0.53) (0.00)
Ethnicity:
Not BAME ref ref ref ref
BAME -0.03∗ 0.05∗ 1.43 -0.03∗
(0.01) (0.02) (0.96) (0.01)
Age:
40-49 ref ref ref ref
20-29 -0.04∗∗ -0.09∗∗∗ -3.19∗∗ -0.03∗∗
(0.01) (0.02) (0.98) (0.01)
30-39 -0.01 -0.05∗ -1.90∗ -0.01
(0.01) (0.02) (0.83) (0.01)
50-59 -0.01 -0.02 -1.30 -0.01
(0.01) (0.02) (0.72) (0.01)
60-65 -0.01 -0.01 -2.57∗ -0.01
(0.01) (0.02) (1.00) (0.01)
Education:
A-level ref ref ref ref
GCSE or lower -0.01 -0.01 -1.02 -0.01
(0.01) (0.02) (0.86) (0.01)
Degree 0.01 0.09∗∗∗ 3.99∗∗∗ 0.00
(0.01) (0.02) (0.70) (0.01)
Household type:
Multiple adult, no children ref ref ref ref
Single adult, no children 0.01 0.03 -0.16 0.01
(0.01) (0.02) (1.00) (0.01)
Single adult, children 0.02 -0.03 -3.40∗ 0.02
(0.02) (0.04) (1.59) (0.02)
Multiple adult, children 0.01 -0.03 -0.80 0.01
(0.01) (0.01) (0.62) (0.01)
Notes: Standard errors in parentheses. Dependent variables are the change from February to April. Earnings are transformed
by the hyperbolic sine function. Employment, hours and earnings include both employees and self-employees. Columns 1-2 are
for all individuals; columns 3-4 refer to those employed in February. Sample sizes (All) are: 10,803 (col 1-2), 8747 (col 3), and
8015 (col 4).* p<0.05, ** p<0.01, *** p<0.001.
25
Figure 1: Employment Changes for Men and Women
probability employed
.4 .5 .6 .7 .8 .9
er er a-
l
a-
l gr ee
lo lo le le ee de
w w ve ve de gr
or or e: e: e: e:
SE SE al al m al
C C m m al m
G G fe fe
e: e:
al al
m m
fe
Employed February Employed April
Positive hours February Positive hours April
Notes: N=10708.
26
Table 4: Reasons for Decline in Hours
Loss of
Employer self-employment
cuts Furloughed business Unemployed Health Caring
All reporting a decline in hours 9.8 43.3 13.8 8.0 7.4 7.2
Gender:
Men 9.6 43.9 17.4 8.0 7.5 5.3
Women 9.9 42.7 10.7 8.1 7.4 8.8
Ethnicity:
Not BAME 9.7 44.2 13.7 7.1 7.2 6.9
BAME 10.7 30.9 15.5 20.9 10.6 11.1
Age:
Age 20-29 9.1 54.0 5.7 14.2 4.2 2.5
Age 30-39 9.7 44.9 13.4 7.1 5.0 15.8
Age 40-49 8.8 40.3 15.7 4.5 8.3 12.1
Age 50-59 11.9 39.9 16.5 6.0 8.9 3.0
Age 60-65 7.3 35.0 19.3 10.7 12.5 1.5
Education:
GCSE or lower 11.6 47.8 14.1 8.3 7.6 4.7
A-level 7.8 50.8 10.9 8.6 9.2 4.5
Degree 9.8 35.3 15.6 7.6 6.2 10.4
Household type:
Single adult, no children 10.6 43.5 15.8 7.9 9.8 1.4
Single adult, children 11.1 46.4 11.0 5.7 3.7 13.2
Multiple adult, no children 9.3 44.7 12.8 9.9 8.3 1.5
Multiple adult, children 10.1 41.3 14.9 6.2 6.3 14.2
Long-run income quintile:
1 10.6 43.1 15.5 11.3 9.4 6.7
2 10.0 49.9 12.9 8.0 7.7 7.2
3 10.9 50.7 11.2 4.6 7.4 6.0
4 8.0 40.1 11.8 8.3 7.0 6.8
5 9.1 30.2 18.5 8.8 5.7 9.5
Worker type:
Fixed hours 11.4 53.8 0.0 7.3 6.2 5.9
Flexible hours 9.5 43.4 0.0 8.7 3.3 13.6
Employer sets (sure min.) 16.0 59.2 0.0 9.0 6.9 3.2
Employer sets (no sure min.) 14.8 58.8 0.0 19.7 6.1 3.8
Self-employed 2.1 7.4 61.0 7.1 12.2 10.8
Works at home (Feb):
Sometimes or always 8.1 25.6 25.7 5.5 5.8 13.8
Never 10.4 50.1 9.3 9.0 8.1 4.7
Notes: Each cell refers to a percentage of those reporting a decline in weekly work hours. Respondents are allowed to report
multiple reasons for an hours decline and so the rows do not sum to one. Sample size (All): 3993.
27
Figure 2: Household Earnings across the Distribution
1000
Household weekly earnings
400 600 800
200
1 2 3 4 5
Long-run income quintile
February April
Notes: N=9341.
28
Table 5: Household Earnings Pre and Post Covid
Full sample Positive February Earnings
Lost 5% Lost 20%
February April February April or more (%) or more (%)
All 538 495 586 530 38 23
Gender:
Men 560 515 607 548 37 23
Women 519 478 567 513 38 24
Age:
Age 20-29 518 482 566 522 40 23
Age 30-39 556 521 585 542 37 21
Age 40-49 568 528 607 551 37 23
Age 50-59 561 512 605 544 36 23
Age 60-65 436 379 525 440 39 31
Ethnicity:
Not BAME 547 503 594 537 38 23
BAME 441 404 495 445 36 27
Education:
GCSE or lower 418 365 469 403 42 27
A-level 487 449 531 484 40 25
Degree 632 591 675 619 34 21
Household type:
Single adult, no children 491 446 616 513 39 32
Single adult, children 327 254 427 308 47 35
Multiple adult, no children 582 536 630 574 37 23
Multiple adult, children 512 476 538 495 37 21
Long-run income quintile:
1 298 255 351 292 44 31
2 394 353 437 380 42 25
3 496 456 530 482 35 21
4 630 583 670 607 36 20
5 832 787 868 813 33 21
Notes: Household earnings are means, weekly, net and equivalised. Columns 1-2 refer to all individuals and
columns 3-6 to individuals in households with positive earnings in February. Sample sizes (All): 9208 (col
1-2), 8511 (col 3-6).
29
Table 6: Distribution of Changes in Household Earnings
p10 p20 p25 p50 p75 p90
Long-run income quintile:
q1 -1.00 -.43 -.32 .00 .00 .00
q2 -.60 -.29 -.20 .00 .00 .00
q3 -.52 -.22 -.16 .00 .00 .02
q4 -.50 -.20 -.14 .00 .00 .02
q5 -.43 -.20 -.12 .00 .00 .02
Notes: The change in household earnings is the proportional change calculated using the Inverse-
Hyperbolic Sine transformation. Earnings are weekly, net and equivalised.
30
Figure 3: Sources of Mitigation
.35 .35
.3 .3
.25 .25
share share
.15 .2 .15 .2
.1 .1
.05 .05
0 0
1 2 3 4 5 1 2 3 4 5
Long-run income quintile Long-run income quintile
Used saving Applied for mortgage holiday Received a transfer from family or friends Applied for Universal Credit
New borrowing New work Used foodbank
(a) Self-Insurance (b) External
Notes: The sample is individuals who experienced a household earnings loss between February and April
2020. Respondents can report multiple methods of mitigation. Sample size: 3345
31
Table 7: Mitigating Earnings Losses
Family
Used New Mortgage Universal transfer Used
savings Borrowed work hol. credit etc. foodbank
All 25.9 7.0 2.7 9.7 7.3 9.9 1.0
Gender:
Men 24.7 5.6 3.0 9.0 7.6 7.6 1.0
Women 26.9 8.4 2.3 10.3 7.1 11.9 1.0
Ethnicity:
Not BAME 25.3 6.4 2.7 9.4 6.9 9.7 1.0
BAME 33.0 15.7 2.7 13.3 12.8 12.0 1.1
Age:
Age 20-29 26.1 7.0 3.1 4.5 9.5 15.8 0.3
Age 30-39 21.6 9.4 3.0 14.7 9.8 11.7 2.0
Age 40-49 24.5 7.6 2.3 15.1 5.8 9.6 0.2
Age 50-59 30.1 7.1 3.1 7.7 6.9 7.8 1.3
Age 60-65 25.8 1.9 1.2 2.9 3.8 3.2 0.9
Education:
GCSE or lower 24.6 6.1 2.4 9.8 8.7 11.8 1.9
A-level 24.5 7.2 3.2 7.9 7.0 11.3 0.7
Degree 27.5 7.4 2.6 10.5 6.7 7.9 0.5
Household type:
Single adult, no children 20.3 5.5 0.7 3.6 10.0 12.1 4.1
Single adult, children 18.4 12.7 2.0 3.3 5.3 25.6 0.0
Multiple adult, no children 28.5 5.4 3.0 5.4 7.1 6.9 0.5
Multiple adult, children 24.5 8.8 2.7 16.2 7.2 11.7 1.1
Long-run income quintile:
1 30.1 8.9 1.8 7.5 11.6 16.4 2.1
2 25.4 11.0 2.1 9.5 10.1 12.2 1.8
3 24.7 4.9 3.3 9.7 6.6 7.7 0.8
4 23.9 5.1 2.1 10.6 4.1 7.2 0.0
5 25.6 5.1 4.1 11.0 4.2 6.0 0.2
Notes: Each cell refers to a percentage of individuals experiencing a household earnings loss between April
and February. Respondents can report multiple methods of mitigation. Sample size: 3345.
32
Figure 4: Spending Reductions by Household Earnings Loss
.8
Share reduced spending
.4 .6
.2
0
no loss 0-5% 5-20% 20%+
Notes: Sample is individuals who report positive household earnings in February 2020. 8,507 individuals.
Figure 5: Behind on Paying Bills and Rent/Mortgage
.2
.15
share
.1
.05
0
1 2 3 4 5
Long-run income quintile
Behind with some or all bills (2017-2018) Behind with some or all bills (April)
Behind with housing payments (2017-2018) Behind with housing payments (April)
Notes: Reported on bills: 10505 individuals. Reported on housing payments: 7870 individuals.
Notes: The COVID question provides more cues on the sorts of bills that may be included.
33
Table 8: Arrears (bills and housing), Hunger and Reduced Spending
Behind with bills Behind with housing
Reduced
2017-18 April 2020 2017-18 April 2020 Hunger spending
All 5.2 7.4 7.5 7.4 4.6 37.1
Gender:
Men 4.1 6.4 7.2 7.0 4.5 36.4
Women 6.1 8.2 7.8 7.7 4.7 37.7
Ethnicity:
Not BAME 4.7 6.6 7.0 6.4 4.4 36.5
BAME 10.6 15.8 12.4 16.7 6.4 43.6
Age:
Age 20-29 5.4 8.4 10.4 10.0 10.4 41.2
Age 30-39 6.7 9.1 7.4 6.6 5.0 37.9
Age 40-49 5.4 8.2 6.8 8.0 3.4 39.0
Age 50-59 5.2 6.5 6.8 6.4 2.8 35.7
Age 60-65 2.4 4.0 6.0 5.1 1.6 30.1
Education:
GCSE or lower 8.1 12.4 10.9 11.6 6.6 39.2
A-level 5.0 6.9 8.4 7.1 5.7 38.9
Degree 3.4 4.5 4.9 4.9 2.7 34.7
Household type:
Single adult, no children 7.3 11.0 11.8 9.4 4.8 29.5
Single adult, children 14.0 19.6 17.7 12.6 5.9 39.5
Multiple adult, no children 3.5 4.7 6.3 6.8 4.5 35.3
Multiple adult, children 6.0 8.8 6.9 7.0 4.5 41.0
Long-run income quintile:
1 12.6 17.8 17.9 15.9 7.7 44.6
2 7.8 11.0 9.7 7.6 7.4 41.6
3 3.1 4.0 5.5 4.3 3.4 36.3
4 2.2 2.7 3.1 4.6 3.1 33.0
5 0.6 2.0 1.0 4.6 1.5 30.3
Notes: Each cell refers to a percentage of the population. Hunger refers to individuals who report a time
last week when they or others in their household were hungry but did not eat. Sample sizes (All): 10,505
(col 1-2), 7870 (col 3-4); 10,617 (col 5); 10,788 (col 6).
34
Table 9: Subjective Financial Situation: Past and Present
Finding it difficult Just about getting by Living comfortably
2017-18 April 2020 2017-18 April 2020 2017-18 April 2020
All 8.4 7.8 23.2 20.3 68.4 71.9
Gender:
Men 7.9 7.7 21.3 20.6 70.8 71.7
Women 8.9 8.0 24.8 20.0 66.3 72.0
Ethnicity:
Not BAME 7.8 7.1 22.5 19.7 69.7 73.2
BAME 14.4 15.5 30.7 26.8 54.9 57.7
Age:
Age 20-29 7.4 7.8 19.9 15.6 72.8 76.5
Age 30-39 9.6 9.2 25.5 19.5 64.9 71.3
Age 40-49 9.7 7.4 25.0 26.0 65.3 66.6
Age 50-59 8.7 8.1 23.6 21.4 67.7 70.5
Age 60-65 5.3 6.1 20.5 15.9 74.1 78.0
Education:
GCSE or lower 11.6 11.5 27.6 28.1 60.7 60.4
A-level 8.4 8.4 23.3 20.2 68.3 71.4
Degree 6.3 5.3 20.5 15.5 73.2 79.2
Household type:
Single adult, no children 13.0 11.6 28.7 24.2 58.3 64.1
Single adult, children 20.5 17.9 39.6 29.5 39.9 52.6
Multiple adult, no children 6.4 6.5 19.7 16.7 73.9 76.8
Multiple adult, children 8.9 7.8 25.1 23.3 66.0 68.9
Long-run income quintile:
1 19.6 15.7 35.6 32.7 44.8 51.6
2 10.0 10.9 31.3 27.1 58.6 62.0
3 6.5 5.5 21.9 18.9 71.6 75.6
4 4.4 4.6 17.9 14.8 77.7 80.6
5 2.0 2.9 10.1 8.6 87.9 88.4
Notes: Each cell refers to a percentage of the population. The categories are derived from a question asking
individuals how they are managing financially. Columns 1-2 refer to individuals ’finding it ’difficult’ or ’very
difficult’; columns 3-4 to those ’just about getting by’; and columns 5-6 to those ’living comfortably’ or
’doing alright’. Sample size (All): 10,732.
35
Table 10: Changes in arrears, financial distress and spending
Behind Behind Financial Reduced
with bills with housing distress spending
Constant 0.00 -0.01 -0.01 0.37∗∗∗
(0.02) (0.04) (0.03) (0.04)
Gender:
Male ref ref ref ref
Women -0.00 0.00 -0.01 0.01
(0.01) (0.01) (0.01) (0.01)
Ethnicity:
Not BAME ref ref ref ref
BAME 0.03 0.05 0.02 0.06∗∗
(0.02) (0.03) (0.02) (0.02)
Age:
40-49 ref ref ref ref
20-29 0.01 -0.02 0.03 0.04
(0.02) (0.03) (0.02) (0.02)
30-39 -0.00 -0.02 0.02 -0.01
(0.01) (0.02) (0.02) (0.02)
50-59 -0.01 -0.02 0.02 -0.01
(0.01) (0.02) (0.01) (0.02)
60-65 -0.01 -0.03 0.03 -0.05∗
(0.01) (0.02) (0.02) (0.02)
Education:
A-level ref ref ref ref
GCSE or lower 0.03 0.02 0.00 0.02
(0.01) (0.02) (0.01) (0.02)
Degree -0.01 0.01 -0.01 -0.04∗
(0.01) (0.01) (0.01) (0.02)
Household type:
Multiple adult, no children ref ref ref ref
Single adult, no children 0.02 -0.03 -0.01 -0.05∗
(0.02) (0.03) (0.02) (0.02)
Single adult, children 0.04 -0.07 -0.02 0.03
(0.04) (0.04) (0.04) (0.04)
Multiple adult, children 0.01 -0.01 -0.01 0.05∗∗
(0.01) (0.02) (0.01) (0.02)
Notes: Standard errors in parentheses. Dependent variables are the change from February to April. We
count as in financial distress those reporting their subjective financial situation as ’finding it very difficult’.
Sample sizes are: 10,393 (col 1), 7772 (col 2), 10,626 (col 3) and 10,674 (col 4).* p<0.05, ** p<0.01, ***
p<0.001.
36
Table 11: Subjective Financial Situation: Future
Next month Next month Next month
Worse off About the same Better off
All 19.7 71.0 9.3
Gender:
Men 19.0 71.1 9.9
Women 20.3 70.9 8.9
Ethnicity:
Not BAME 19.2 71.5 9.3
BAME 24.7 65.5 9.8
Age:
Age 20-29 16.0 68.1 15.9
Age 30-39 20.5 67.6 11.8
Age 40-49 22.1 69.3 8.6
Age 50-59 20.3 73.6 6.1
Age 60-65 18.2 77.2 4.6
Education:
GCSE or lower 21.6 71.3 7.1
A-level 19.2 70.3 10.5
Degree 18.8 71.2 10.0
Household type:
Single adult, no children 16.7 75.0 8.4
Single adult, children 25.8 67.3 6.9
Multiple adult, no children 18.2 72.4 9.4
Multiple adult, children 21.9 68.5 9.6
Long-run income quintile:
1 24.0 67.9 8.1
2 23.4 67.9 8.8
3 18.3 73.3 8.4
4 16.4 72.1 11.5
5 16.8 73.5 9.8
Notes: Each cell refers to a percentage of the population. Sample size (All): 10,783.
37
Table 12: Statistical Testing of Survey Weights
Wave 9 Covid Test Statistic
Weighted Unweighted Basic weight Refined weight Basic Refined
In refined weights only:
Subjective financial situation:
Living comfortably/doing alright 0.71*** 0.76*** 0.76*** 0.72*** -0.04*** -0.00
(0.000) (0.000) (0.000) (0.000) (0.000) (0.557)
Just about getting by 0.21*** 0.18*** 0.18*** 0.21*** 0.03*** 0.00
(0.000) (0.000) (0.000) (0.000) (0.000) (0.382)
Finding it quite/very difficult 0.07*** 0.06*** 0.06*** 0.07*** 0.01*** 0.00
(0.000) (0.000) (0.000) (0.000) (0.000) (0.901)
Housing tenure:
Owned 0.65*** 0.59*** 0.59*** 0.65*** 0.06*** -0.00
(0.000) (0.000) (0.000) (0.000) (0.000) (0.932)
Rented or mortgage 0.35*** 0.41*** 0.41*** 0.35*** -0.06*** 0.00
(0.000) (0.000) (0.000) (0.000) (0.000) (0.889)
In neither weighting model:
Mental component summary 48.25*** 49.06*** 48.88*** 48.27*** -0.62 -0.01
(0.000) (0.000) (0.000) (0.000) (0.134) (0.972)
Household net income 3365.95*** 3706.34*** 3583.01*** 3546.84*** -217.05*** -180.89***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Receives core benefit 0.05*** 0.03*** 0.03*** 0.04*** 0.02*** 0.01**
(0.000) (0.000) (0.000) (0.000) (0.000) (0.006)
Behind with council tax 0.06 0.03 0.04 0.05 0.02 0.01
(0.000) (0.000) (0.000) (0.000) (0.000) (0.003)
Notes: ‘Core benefits’ include Income Support, Job Seeker’s Allowance and Universal Credit. ‘Mental component summary’
is a validated mental health functioning score. It is derived from the short form 12-item Survey (SF-12) and takes values from
0-100 with a higher score indicating better functioning.
38
Appendix
Table 13: Summary Statistics
Unweighted Weighted
Men 0.400 0.456
BAME 0.132 0.086
Age 46.346 44.611
Highest Qualification:
GCSE or lower 0.240 0.287
A-level 0.226 0.253
Degree 0.534 0.460
Household type:
Single Adult, no children 0.083 0.090
Single adult, children 0.031 0.032
Multiple adult, no children 0.484 0.494
Multiple adult, children 0.402 0.384
Worker type:
Fixed hours 0.676 0.670
Flexible hours 0.083 0.075
Employer sets (sure min.) 0.068 0.079
Employer sets (no sure min.) 0.024 0.029
Self-employed 0.150 0.147
Works at home:
Sometimes or always 0.341 0.307
Never 0.659 0.693
Occupation:
Elementary 0.069 0.090
Skilled trades 0.049 0.071
Sales, customer service 0.069 0.090
Process, plant, machine operatives 0.046 0.051
Caring, leisure, other service 0.094 0.098
Administrative, secretarial 0.130 0.120
Managers, directors 0.112 0.114
Associate professional, technical 0.185 0.165
Professional 0.245 0.200
Notes: 10,803 individuals aged between 20-65. Minimal item non-response. Occupation is recorded at wave 9 (2017-18).
39
Table 14: Shocks by industry
(1) (2) (3) (4) (5)
Positive Positive
Employed hours hours Hours Hours
(April) (Feb) (April) (Feb) (April)
Accommodation, Food Service 0.91 0.99 0.31 33.94 9.32
(0.03) (0.01) (0.04) (1.06) (1.21)
Arts, other service 0.97 1.00 0.52 33.28 15.63
(0.01) (0.00) (0.04) (0.93) (1.44)
Construction, Real Estate 0.96 1.00 0.56 38.90 18.81
(0.01) (0.00) (0.03) (0.71) (1.26)
Wholesale, Retail Trade 0.96 0.99 0.60 33.79 19.91
(0.01) (0.00) (0.02) (0.56) (0.86)
Transportation, Storage 0.98 1.00 0.64 39.27 24.48
(0.01) (0.00) (0.04) (0.72) (1.51)
Administrative, Support Service 0.95 0.97 0.66 33.46 20.62
(0.02) (0.01) (0.04) (1.05) (1.51)
Manufacturing, Agriculture, 0.97 0.99 0.67 39.29 25.23
Mining, Utilities (0.01) (0.00) (0.02) (0.41) (0.98)
Education 0.97 0.99 0.73 33.02 20.18
(0.01) (0.00) (0.02) (0.56) (0.69)
Professional, Scientific and Technical 0.99 1.00 0.79 35.64 27.47
(0.01) (0.00) (0.02) (0.47) (0.97)
Information, Communication 0.95 0.99 0.80 36.45 29.54
(0.02) (0.00) (0.03) (0.60) (1.38)
Human Health, Social Work 0.98 0.98 0.82 32.59 27.67
(0.01) (0.00) (0.01) (0.41) (0.64)
Public Administration, Defence 0.98 0.99 0.86 34.89 29.67
(0.01) (0.01) (0.02) (0.47) (0.81)
Financial, Insurance 0.98 0.99 0.87 36.77 32.45
(0.02) (0.00) (0.03) (0.68) (1.20)
Notes: Sample of 7834 individuals who reported an industry at wave 9 (2017-18).
Standard errors in parenthesis.
40
File and source
- File
- 2020-the-idiosyncratic-impact-of-an-aggregate-shock-the-distributional-3-consequences-of-covid-19-the-ifs-http-dx-doi-or.pdf
- Size
- 650,262 bytes
- SHA-256
- 8bf6fe1e9600145f407839c509ec4138d1ffadf7d02d7411b06be1fd90b86e74
- Original
- www.econstor.eu